Roadmap: Path to v1.0 (Dual-Engine & Causal Runtime)

[akarlaraytu]

Strategic Roadmap: From Architecture to Adoption

This roadmap tracks the evolution of csl-core towards a stable v1.0 release.

Current Status:
The "Single-Engine" (Z3) architecture is fully implemented and production-ready. We have functional modules for:

• Engines: Z3 SMT Solver (chimera_core.engines.z3_engine)
• Runtime: AST Parser & Compiler (chimera_core.language)
• Integration: LangChain Plugins (chimera_core.plugins)
• Interface: CLI Tools (chimera_core.cli)
• Documentation: Comprehensive guides available in docs/ and quickstart/.

Our goal for v1.0 is to evolve into a "Dual-Engine" runtime (Z3 + TLA+) and eventually integrate Causal Reasoning.

---

✅ Phase 1: The Foundation (Completed)

We have laid the groundwork for deterministic AI safety.

• CSL Compiler: Custom EBNF grammar, tokenizer, and AST parser (chimera_core.language).
• Z3 Engine: SMT-based constraint solver for mathematically proving policy compliance (engines.z3_engine).
• Agent Plugins: Native BaseTool wrapper for LangChain (plugins.langchain).
• Developer CLI: Comprehensive command-line interface for CI/CD pipelines (cli.py).
• Documentation: Zero-to-hero guides and example policies (docs/).

🚧 Phase 2: Temporal Verification (Current Focus - Q1 2026)

Adding the dimension of "Time" and "State" to safety checks.

• TLA+ Engine Integration: Implement the engines.tla_engine module to handle concurrent state verification.
• Auto-Transpilation: Create a transpiler to convert CSL policies directly into .tla specifications (removing the need to write raw TLA+).
• State-Space Exploration: Enable the runtime to verify sequences of actions (e.g., "User cannot withdraw twice in the same block").
• Dual-Engine Benchmarking: Optimize the switching logic between Z3 and TLA+ to ensure <15ms latency per decision.

🌍 Phase 3: Ecosystem & Scale (Q2 2026)

Making CSL ubiquitous and easy to deploy.

• WASM Compilation: Compile the CSL Validator to WebAssembly to enable client-side verification in browsers and edge devices (removing Python dependency).
• Distributed Verification: Architect the engine to support off-chain proof generation, making it compatible with decentralized node networks in the future.
• Standard Policy Library: A centralized registry (like NPM) for pre-verified, open-source CSL policies (e.g., "Standard Treasury Guard").

🧠 Phase 4: Causal Intelligence (Q3 2026)

Moving from "What happened?" to "Why did it happen?".

• Counterfactual Simulation: Allow agents to ask "If I execute this trade, what's my 95% CI on portfolio risk?"
• Intervention Analysis: Enable "What-If" scenario testing (Do-Calculus) before execution commitments.
• Causal Audit Trails: Upgrade logs to provide causal graphs for blocked actions.
  • Example: "Transfer blocked → caused by (risk_score > 80) ← caused by (3 failed auth attempts)"

---

🤝 Contributing

We are currently heads-down building Phase 2 (TLA+ Integration).

Since the development on the core goes fast, we need more real-world use cases and examples. Please feel free to use CSL, break it, and create use cases to publish. Policies from any domain are welcome!

See good first issue & policy to get started.